Welcome to lightsaber

Overview

lightsaber is designed ground up to provide a simple, composible, and unified model training framework. It has been designed based on state-of-the-art open source tools and extended to support the common use cases for disease progression modeling (DPM).

lightsaber provides four main components:

• Data ingestion modules
• Model Trainers
• DPM problem specific model evaluation
• Model tracking and support for post-hoc model evaluation.

Each of these components are designed such that a user should be able to pick some or all of the modules and embed these seamlessly with their current workflow. Futhermore, when used in the recommended manner, lightsaber provides a batteries included approach allowing the modeler to focus only on developing the logic of their model and letting lightsaber handle the rest.

Currently, we support the following DPM use cases:

• classification: one or multi-class

Also, we support and extend the following frameworks:

• scikit-learn compliant models: for classical models
• pytorch compliant models: for general purpose models, including deep learning models.

To summarize, it is thus an opinionated take on how DPM should be conducted providing with a unified core to abstract and standardize out the engineering, evaluation, model training, and model tracking to support: (a) reproducible research, (b) accelarate model development, and (c) standardize model deployment.

Installation Instructions

Lightsaber is installable as a python package.

It can be installed using conda as:

conda install -c conda-forge dpm360-lightsaber

or from pypi as:

pip install dpm360-lightsaber

It can also be installed from source using pip as follows:

• barebones install of Lightsaber: pip install .
• with doc support: pip install .[doc]
• with time-to-event modeling (T2E) support: pip install .[t2e]
• full install with all components: pip install .[full]